Stochastic Dual Coordinate Ascent Methods for Regularized Loss Minimization thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Stochastic Dual Coordinate Ascent Methods for Regularized Loss Minimization

Published on Jan 16, 20133673 Views

Stochastic Gradient Descent (SGD) has become popular for solving large scale supervised machine learning optimization problems such as SVM, due to their strong theoretical guarantees. While the closel

Related categories

Chapter list

Stochastic Dual Coordinate Ascent Methods for Regularized Loss Minimization00:00
Regularized Loss Minimization (1)00:07
Regularized Loss Minimization (2)00:28
Dual Coordinate Ascent (DCA)00:58
SDCA vs. SGD - update rule02:16
SDCA vs. SGD - update rule - Example03:10
SDCA vs. SGD - experimental observations (1)03:33
SDCA vs. SGD - experimental observations (2)04:23
SDCA vs. SGD - Current analysis is unsatisfactory04:45
Dual vs. Primal sub-optimality06:24
Our results08:16
SDCA vs. DCA - Randomization is crucial10:37
Smoothing the hinge-loss (1)11:32
Smoothing the hinge-loss (2)12:15
Smoothing the hinge-loss (3)13:26
Additional related work13:43
SDCA vs. DCA - Randomization is crucial16:07
Extensions16:33
Proof Idea (1)17:13
Proof Idea (2)17:16
Summary17:17